Research On Optimal Microcirculation System Of Community Shuttle Linked With Metro Service

As one of the most important modes of urban public transport, the metro has many advantages compared to the ground transit, such as high-speed, reliable, and it is commonly used by passengers in big cities nowadays. On the other hand, as another irreplacable mode in the urban public transport, the community shuttle has flexible features and could overcome some insurmountable shortcomings of metros, such as poor accessibility and failing to supply the convenient door-to-door services. Based on the complementary relationship of the two transit modes, this research aims at developing an efficient and robust microcirculation system of community shuttles linked with metro service from the following aspects:optimal routing design, coordinated timetable development, fleet scheduling and dispatching based on the demand-responsive service. Based on a good trade-off between passengers and transit operators, the optimization of public transit microcirculation could reduce the travel costs of passengers while ensuring a reasonable investment of operators, thus enhancing the whole level of service in the multi-mode transit system. Main contributions of the dissertation are as follows:(1) Developing two types of routing optimization models for community shuttles. The first one is based on a realistic network. In the model formulation, a potential bus demand indicator from the view of segments is defined first, and then the model is established with the objective of maximizing the defined demand indicator, considering the maximum bus travel time. The second model is formulated based on a semi-realistic network, with the objective of minimizing the total cost (including user and supplier costs). To calculatethe user cost, a heuristic algorithm for locating stops is presented, and then the optimized headway corresponding to a given bus route is confirmed by minimizing the total cost function. In solving the two problems, a depth-first search algorithm (DFS) is developed first for exhaustively searching all the solutions; and then an improved genetic algorithm (GA) is proposed. For validating the two algorithms, a real-life and a simulated example are presented. By comparing the results and the CPU time of the two algorithms, the research found that GA is reliable and efficient in solving the two problems. The impacts of route length and the maximum tolerable walking distance on the related costs and headways are also analyzed.(2) Developing a coordinated timetable optimization model, with the objective of minimizing passenger’s travel cost (including schedule delay and transfer costs) when the number of vehicle trips and the fleet size are both given. Two constraints, i.e. the vehicle capacity and fleet size, are considered in the model. The first constraint is treated as soft, and it is reflected by adding an overloading penalty in the original objective function; while the latter one is handled by a proposed timetable generating method. Two algorithms are employed to solve the problem, i.e. genetic algorithm and a Frank-Wolfe algorithm combined with a heuristic algorithm of shifting departure times (FW-SDT). The simulated and real-life examples confirm the feasibility of the two algorithms, and demonstrate that FW-SDT outperforms GA in both accuracy and effectiveness.(3) Introducing another type of transit, demand-responsie transit (DRT), into the microcirculation system of community shuttles linked with metro service. This type of transit is more flexible compared to the conventional ground transit. Fleet scheduling and dispatching strategies of community shuttles are optimized through a model based on DRT. The model is formulated with the objective of minimizing the total cost (including operation and passenger’s in-vehicle costs), considering a series of realistic constraints, such as time window, passenger’s in-vehicle time, vehicle capacity and maximum travel time of the vehicle. In solving the problem, two algorithms are employed, i.e. tabu search (TS) and variable neighborhood search based simulated annealing (VNS-SA). A numerical example based on realistic network is presented to validate the two algorithms. For making a resonable trade-off between optimization results and CPU time, the two algorithms with different internal algorithm combinations are used to solve the problem respectively, and some comparsion of the related results follows.